ISPRS SIPT IGU UCI CIG ACSG Table of contents Table des matières Authors index Index des auteurs Search Recherches Exit Sortir SEMANTIC SIMILARITY EVALUATION MODEL IN CATEGORICAL DATABASE GENERALIZATION 1 2 Liu Yaolin , Martin Molenaar , Menno-Jan Kraak 2 1 School of Resource and Environmental Science Wuhan University, Wuhan, 430070, P.R. China 2 Department of Geo-Informatics International Institute for Aerospace Survey and Earth Science 7500 AA, Enschede,The Netherlands Email {yaolin, Molenaar, Kraak}@itc.nl} Commission IV, WG IV/3 KEY WORDS: Hierarchical structure, Similarity Model, Categorical Database, Classification System, ABSTRACT Database generalization process will be used to derive a new database with less detail for some application purposes from a single detailed database. In a database generalization process, semantic similarity measures among objects and among object types play a key role in object aggregation process. In this paper a generalization procedure will be developed based on object aggregation. The decision to aggregation objects will be based on the geometric properties of the objects, on their spatial relationships and on their thematic similarity. Normally, this similarity among objects acts as decisive rule that controls the generalization operations. This paper focuses on semantic similarity evaluation model for categorical database generalization. It presents a semantic evaluation model after reviewing current similarity models. The models are based on classification hierarchies, set theory, and attribute structure of classes. It is a computation model and can not only be used to compute the similarity between objects at same level, or at different levels, but also the similarity between object types at the same level and at different level. 1. INTRODUCTION set theory, hierarchy structure and attribute structure of class. The similarity measures among objects and among object types play a key role in object aggregation in database generalization. The aggregating two objects not only depends on the geometric properties of the two objects, but also the thematic The paper is organized as following. First, the brief review is given; secondly, the basic concepts for similarity measure are given and followed by the computing similarity model is proposed; finally an example is used to test the model. properties of the two objects. Using Set theory, Tversky (1977) defines a similarity measure in terms of matching process. This measure produces a similarity value that is not only the result 2. BASIC CONCEPTS of the common, but also the result of the different Some basic concepts must be clear before discussing similarity computing model. characteristics between objects, which is in agreement to an 2.1 Class, Object Type and Object information theory definition of similarity (Lin,D 1998, Rodriquez and Egenhofer 1999, Bishr, 1997, Chakroun et In this paper, the class and object type have the same meaning. A class or object type determines a set of attributes to form its al,2000, Rodríguez and M. Egenhofer, 1999, Rodríguez, M. attribute structure. Each class c j or object type c Egenhofer and Rugg, 1999). Although many models of own attribute structure List (c j ) as following: similarity are defined in the literature, the similarity measure method based on set theory, hierarchy structure and attribute structure of class is still lack. This paper mainly discusses the similarity measure method in the database generalization used j has its List(c j )={A 1 ,…A i ,…A n } A i denotes one of the attributes of class c j . Each attribute will have a name, a domain that will be specified by defining Symposium on Geospatial Theory, Processing and Applications, Symposium sur la théorie, les traitements et les applications des données Géospatiales, Ottawa 2002 the range of the attribute values and scale type of the domain which indicates whether these values are from a nominal, an ordinal, an interval or ratio scale. An attribute A i can be specified by a three tuple (Molenaar 1998): However there are still lack of formalization of classification hierarchy and aggregation hierarchy based on Set theory and properties of class. In the following part, the formalizations of classification A i ={NAME(A i ), SCALETYPE(A i ), DOMAIN(A i )} For each object which belong to class c j , the attribute hierarchy and aggregation hierarchy are discussed. The object types in a geo-spatial model are normally determined by classification hierarchy and aggregation structure defined by a class specifies the description structure hierarchy. Changing classification hierarchy and aggregation of the object. A value is assigned to every attribute in the hierarchy will results in changing geo-spatial model associated attribute structure of object. For any object, its direct class is with database, and in turn changing the contents of the database. unique and lowest in a classification hierarchy since different Changing the attribute structure and extension of classes at classes have different attribute structures. These values must different level in classification hierarchy and aggregation fall within the range of the attribute domain, which must be hierarchy associated with a database will induce a new defined prior to the actual assignment of attribute values. classification hierarchy and aggregation hierarchy and define a new data model of database and rebuild the corresponding Class A1 A2 … An contents of database. Let S be the set of objects {o 1 ,o 2 , o 3 ,…,o i }in space, U be Object i Figure 1 a1 a2 … an the set of classes {c 1 ,c 2 ,c 3 ,…,c j } for the database space. Before formalizing classification and aggregation hierarchy, the relations between classes must be identified. Diagram representing the relations between objects, classes and attributes ( after Molenaar 1998 ) Let c i , c j ∈ U be two arbitrary spatial classes, there should be no objects that belong to the extensions of the two different Each object has an attribute structure list containing one value classes of U. Based on the definition of classification hierarchy a j for every attribute of its class. The thematic description of last chapter and relations among classes, we can formalize an object can now be specified by its class (which specifies the attributes of the object) together with the list of attribute values. classification hierarchy with intension and extension of the 2.2 classes. Intension and Extension of A Class • c ψ i c Ext(c i ) ∩ Ext(c the value of a combination of attributes. The condition specifies a subset of the set containing all the possible combinations of among classes symbolized by the values of the attributes, denoted as Int(c j ), while the set of with c j ); all the objects that belong to c j with the same attribute structure is commonly identified as the extension of the class • ≡ ci Ext(c j ) of c Hierarchy • ⊆ ci ≤c j ), (c i is consistent abstraction and generalization (Smith and Smith 1977, symbolized by j ) of c is equal to i and c i is identical to c j ); c j ;( Relations of inclusion among classes ci of c ≡ ),(Ext(c i c j , only if Ext(c i ) ⊆ Ext(c j ), denoted as databases and knowledge bases, especially in relation to data Yee,Leung, Kwong,S.L. nad He J.Z 1999, Molenaar 1996). ψ c j , only if Ext(c i )=Ext(c j ), denoted as symbolized by Formalizing Classification Hierarchy and Aggregation The class hierarchy has been studied for many years in if c i = c j ; (Relations of equivalence among classes c j , denoted as Ext(c j ). 2.3 only ) ≠ φ ;( Relations of consistent The intension of a class c j can be expressed as a condition for j , j ⊆ ), (Ext(c i ) of c i and c i belongs to c j ) include Ext(c j ) or • ⊂ ci c j , if Ext(c i ) ⊂ Ext(c j ), denoted as c i <c c j . We also call c i a sub-class of c and c j a j super-class of c i . (Relations of complete inclusion among classes symbolized by ⊂ c i completely include Ext(c ) of c j the properties of the hierarchical structure. A, B and C in Figure 2 represent the different branches in the hierarchical structure. For later use, they are called sub tree. T in the same Figure is called the top of the structure and c i are object or object type. ), (Ext(c i ) of j and c T i completely belongs to c j ). • c φ ≠ i c j T , only if Ext(c i ) ∩ Ext(c j )= g .( Relations of disjunction among classes symbolized by ≠ ), between c i and c j , and c i is different from c j ). j if and only if the extension of c i is subsumed by the extension of c9 c6 c5 c10 c8 c17 Example of a hierarchical structure Figure 2 is a partial binary relation on U, called the `Belong to’ relation, and (U, c7 c15 c j . We call c i ‘IS-A’ c j . ≤c C c14 (there is no common object A class c i is included by a class c Obviously, B A ≤ c) is a partially ordered set that The semantic similarity matrix represents the similarity between object types. we call a classification hierarchy. The process of formalization of classification hierarchy depicts how the object types (classes) Table 1 Example of semantic similarity matrix and super object types can be formed into a hierarchical structure. For creation of a new super object type in the classification hierarchy there will be: SIMILARIT Y Sub-type1 Sub-type1 If any A, B ∈ H and no D ∈ H satisfying Ext(D) =Ext(A) ∪ Ext(B), ∪ … type2 11 s 14 s 15 s 24 s 25 Sup-type1 s 17 s 27 … … … … … … … … s s 44 45 s 47 s 57 … type2 Ext(B), LIST(D) =LIST(A) ∩ LIST(B) and D ∈ H, … … type1 then generate such a class D, and let Ext(D) =Ext(A) Sub-type2 type1 … s • … s 55 … … … Sup-type1 noted as IS-A links; s 77 … … … … … … … … … … The upward connections from objects to classes and classes to super classes are is-a links, which express that an object is an instantiation of a class and that a class is a special case of more Where: sub-type1 etc denote different elementary object types; general super-class. At each level, the classes inherit the type1,type2 denote different object types; attribute structure of their super-classes at the next higher level sup-type1etc denote (super) composite object type; and propagate it normally with an extension to the next lower level. At lowest level in the hierarchy are elementary objects (Molenaar 1998). 2.4 Hierarchic Semantic Similarity Matrix For a hierarchical structure as shown in Figure 2. a semantic similarity matrix as shown in Table 1 can be defined based on s ij denotes similarity value among object types. The larger the value of an element in the matrix is, the more the similarity between two object types that the element links is. The matrix is symmetric and reflexive one, and has the property that s ij is equal to s ji (s ij = s ji ) and s ii is equal to s jj (s ii = s jj =1) in the matrix. s ji is a value between o and 1. This matrix shows the similarity among different levels of object types. It will provide potential possibility to chose objects of different types to be merged or aggregated. The similarity matrix will be used as a look-up table for guiding or governing the aggregation process of spatial objects in and immediate super object types that subsumes them which reflects difference properties between two given object types and the other is the distance between immediate super object types that subsumes two given object types and the top of hierarchical structure which reflects the common properties of two given object types. For the second case, the distance between immediate super object types that subsumes two given object types and the top of hierarchical structure will be zero semantics to a certain application. The value of element s ij in the matrix can be given by expert knowledge or by calculation (to be discussed in Section 6.3.2) based on aggregation hierarchy and classification hierarchy. since the two given object types belong to different sub tree such as c 1 and c 15 in Figure 2. So this distance will be replaced by the correlation value between two sub trees in the 3. COMPUTING MODEL OF SIMILARITY Equation 2. A computational model that assesses similarity among objects and object types based on set theory, classification hierarchy and attribute structure of class is proposed. There are three distances. One is the distance (number of the link edges) from immediate super object type that subsumes c i and c j s ij (c i to the top of a hierarchy such as object type g to top of the tree T in l l + α (c , c ) * d + (1 − α (c , c )) * d i j ci i j cj ,c j )= β ( , ) * (1 − α (ci , c j )) * d cj + + β α c c d i j ci (1) Figure 2 which represents common part of attribute structure between two object types c i and c j . Another distance (number of the link edges) from immediate super object type that subsumes c i and c j to c i such as object type g to c 1 in j (c i and c j ∈ same sub-tree) ∉ different sub-tree) (b) immediate super object type that subsumes c i and c d ci : the shortest distance (number of the link edges) from immediate super object type super object type that subsumes c i and c c to c j such as d cj : object type g to c 5 in Figure 2 which represents the different part of attribute structures between object type c i and c j (| c j belonging to the same sub tree such as sub tree A in Figure 2 and the other is for two given object types belonging to two different sub tree such as A and B as shown in Figure 2. For the j that subsumes c i and to c i ; the shortest distance (number of the link edges) from immediate super object type that subsumes c i and c α: j to c j ; a function of the distance (number of the link edge ) between immediate super object type that subsumes c i and c j to the class c i and c j . - c i |), The proposed model is shown in Equation 2. It suits for two cases. One is for two given objects or object types j to the top of a hierarchy; third is the distance (number of the link edges) from immediate j (a) Where: l: the shortest distance (number of the link edge ) from Figure 2 which represents the different part of attribute structure between object type c i and c j (| c i - c j |). And the (c i and c β : correlation degree among different sub-trees, such as similarity among agriculture land use, forest land use and building up land use, and its value can be given by experts based on application requirement. first case, the model uses two types of distances to define the A natural approach to comparing the degree of generalization common and difference properties between the given object between object types is to determine the distances from these types. One is the distance between given objects or object types object types to the immediate super object types that subsumes them in a classification hierarchy as shown in Figure 1, that is, extreme value 1 represents the case that the two entity classes their least upper bound in a partially ordered set. In a sense, the are completely the same, whereas the value 0 occurs when the difference in the distances from these object types to the two entity classes are completely different. immediate super object types that subsumes them in a classification hierarchy reflects the difference in attribute structure between two object types. The α (c i , c j ) can be 4. EXAMPLE An example for computing element of similarity matrix from classification hierarchy is as following: Taking Figure 3 as expressed as a function of the distance d ci and d cj . In order to example to calculate the similarity among object types. The α , the function (Equation (2)) is defined as class code can be seen in appendix. The correlation value get final values of following: among construction land, agriculture land and unused land that is given by the experts is 0.5. α d ci d + d c j (c i , c j )= c i d ci 1 − d ci + d (d ci ≤ d cj ) Land u se (2) C o n s t r u c tio n (d c i >d cj ) c j R e s id e n T r a n s f o r …C u lt iv a t e F o r ti l t ti d dl l d d t where: d ci : P ast … W a t san … d the shortest distance (number of the link edges) from immediate super object type c d cj : U n u sed A g r ic u lt u r e j that subsumes c i and t o w … R a il r o a …I r . d il D ry l d …W o o …N a t . dl d … to c i ; the shortest distance (number of the link edges) from immediate super object type that subsumes c i and c j to c j . Figure 3 Land use classification hierarchy before generalization The computing result of similarity among object types based on the model (see in Table 2.) This similarity function yields values between 0and 1. The 5. CONCLUSIONS The computing model that has been proposed has taken set theory, classification hierarchy and attribute structure into account. It can be used to measure the semantic similarity among object types in classification hierarchy. The example shows that the computing result of the model are reasonable and efficient. How to decide the correlation parameter β will still need to research in the future work. The authors would thanks the financial support from ITC, Ministry of Education,P.R. China and State Bureau of Surveying and Mapping. 6. REFERENCES Bishr, Y., 1997, Semantic aspects of interoperable GIS. ITC, Publication No. 56, Enschede: International Institute for Areospace Survey and Earth Sciences (ITC). Chakroun,H., Benie,G.B., Oneill,N.T., and Desilets,J. 2000, Spatial analysis weighting algorithm using Voronoi diagrams. International Journal of Information Science, Vol.14,No.4 pp.319-336. Lin, D. 1998,An Information theoretic definition of similarity (eds.) International conference on amchine learning, ICML’98. Molenaar, M., 1998,An Introduction to the Theory of Spatial Object Modelling for GIS, Taylor & Francis Ltd. London. Molenaar, M.,1996, The Role of Topologic and Hierarchical Spatial Object Models in Database Generalization, In Molenaar,M.(Eds.) Methods for the Generalization of Geo-databases, Delft: Netherlands Geodetic Commission, New Series, Nr 43,pp.13-36. Rodríguez and M. Egenhofer, 1999, Putting similarity assessment into context: matching functions with the user’s intended operations. Modeling and Using Context, CONTEXT'99, Trento, Italy. In: P. Bouquet, L. Serafini, Patrick Brézillon, Massimo Benerecetti, and Francesca Castellani (eds.), Lecture Notes in Computer Science, Vol. 1688, Spiringer-Verlag, pp. 310-323,. Rodríguez, M. Egenhofer, and R. Rugg , 1999, Assessing semantic similarity among geospatial feature class definition Interoperating Geographic Information Systems, Second International Conference, INTEROP'99, Zurich, Zwitzerland. In: . Vckovski, K. Brassel, and H.-J. Schek (eds.), Lecture Notes in Computer Science, Vol. 1580, Springer-Verlag, pp. 189-202,. 124 125 13 Smith, J.M. and Smith, D.C.P, 1977, Database abstractions: aggregation and generalization, ACM Transactions on Database System, 2, pp.105-133 15 Tversky,A. 1977, Features of similarity. Psychological review, 84(40:PP.327-352). Appendix Land use classification (code) 1. Agriculture Land 11 12 (100) 2. Cultivated land Rubber plantation Other Forest 131 Wood land 132 Shrubbery land 134 Young forestation land 135 Slashes Water area 151 Rivers 153 Reservoir 154 Pond 156 Beaches and flats 158 Hydraulic building Construction Land (200) 21 Residential quarters and industrial and mining land 111 Irrigated paddy fields 112 Rain fed paddy fields 211 113 Irrigated land 212 Residential quarters in rural areas 114 Dry land 213 Isolated industrial and mining 115 Vegetable plots 121 land 3. Garden Land Orchards 122 Mulberry fields 123 Tea fields Area of cities and towns Unused lands (300) 310 Waste lands 380 Others Similarity Table Table 2 S 111 112 113 114 115 121 122 123 124 125 131 132 134 135 151 153 154 156 158 211 212 213 110 120 130 150 210 310 380 100 200 300 111 1 0.7 0.7 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 1 112 0.7 1 0.7 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 1 113 0.7 0.7 1 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 1 114 0.7 0.7 0.7 1 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 1 115 0.7 0.7 0.7 0.7 1 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 1 121 0.3 0.3 0.3 0.3 0.3 1 0.4 0.4 0.4 0.2 0.1 0.1 1 0.2 0.1 0.4 0.4 0.4 0.2 0.1 0.1 1 0.2 0.1 0.2 0.1 0.4 0.4 0.2 0.1 0.1 1 0.2 0.1 0.4 0.4 0.2 0.1 0.1 1 0.2 0.1 0.4 0.2 0.1 0.1 1 0.2 0.1 0.4 0.2 0.1 0.1 1 0.2 0.1 0.2 0.1 0.2 0.1 0.1 1 0.2 0.1 0.2 0.1 0.1 1 0.2 0.1 1 1 0.1 0.1 0.3 1 0.2 0.1 0.1 0.3 1 0.2 0.1 0.1 0.3 1 0.2 0.5 0.2 0.2 0.2 1 0.2 0.2 0.2 0.5 0.5 0.5 1 210 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 1 0.2 0.1 0.1 1 0.5 0.5 0.2 0.2 0.2 1 0.4 0.4 0.4 0.4 0.4 0.2 0.2 0.2 0.5 0.5 1 1 0.2 0.1 0.5 0.5 0.5 0.2 0.2 0.2 1 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.2 0.2 0.2 0.5 1 1 0.2 0.1 0.1 1 0.2 0.2 0.2 0.2 1 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.2 0.2 0.2 1 1 0.2 0.1 0.7 0.2 0.2 0.2 0.2 1 213 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.7 0.7 1 1 0.2 0.1 0.1 1 0.7 0.7 0.2 0.2 0.2 0.2 1 212 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.7 1 150 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 1 0.2 0.1 0.2 0.2 0.2 0.4 0.4 0.4 1 211 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 1 1 0.4 0.2 0.1 0.1 1 0.7 0.2 0.2 0.2 0.4 0.4 0.4 1 158 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 0.7 0.7 0.7 1 1 0.2 0.1 0.7 0.7 0.2 0.2 0.2 0.4 0.4 0.4 1 156 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 0.7 0.7 1 1 0.4 0.2 0.1 0.1 1 0.7 0.7 0.7 0.2 0.2 0.2 0.4 0.4 0.4 1 154 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 0.7 1 130 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 1 0.4 0.4 0.2 0.1 0.1 1 0.7 0.7 0.7 0.7 0.2 0.2 0.2 0.4 0.4 0.4 1 153 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 1 1 0.2 0.1 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 0.4 1 151 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 1 1 0.4 0.4 0.2 0.1 0.1 1 0.7 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 0.4 1 135 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 0.7 0.7 1 1 0.2 0.1 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 0.4 1 134 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 0.7 1 1 0.4 0.4 0.2 0.1 0.1 1 0.7 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 0.4 1 132 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7 1 120 0.4 0.4 0.4 0.4 0.4 1 0.2 0.1 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 1 131 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 1 1 0.4 0.4 0.4 0.2 0.1 0.1 1 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 1 125 0.3 0.3 0.3 0.3 0.3 0.7 0.7 0.7 0.7 1 1 0.2 0.1 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 1 124 0.3 0.3 0.3 0.3 0.3 0.7 0.7 0.7 1 1 0.4 0.4 0.4 0.2 0.1 0.1 1 0.7 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 1 123 0.3 0.3 0.3 0.3 0.3 0.7 0.7 1 1 0.2 0.1 0.7 0.7 0.7 0.7 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.4 1 122 0.3 0.3 0.3 0.3 0.3 0.7 1 110 1 0.4 0.4 0.4 0.2 0.1 0.1 1 0.2 0.2 0.2 1 0.2 0.2 0.2 0.2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0.3 0.3 0.3 1 200 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 1 1 1 1 1 1 0.4 0.1 0.4 0.3 0.2 0.2 0.4 1 300 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.2 0.2 0.2 0.1 0.1 0.1 0.1 0.2 1 0.2 0.2 0.2 1 0.3 0.2 0.2 1 0.4 0.4 0.4 0.4 1 0.4 0.1 0.5 0.2 0.2 1 380 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.2 0.2 0.2 0.2 0.2 0.5 1 1 0.4 0.1 0.2 0.2 0.3 1 310 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.2 0.2 0.2 0.2 0.2 1 100 1 0.4 0.1 1 0.3 0.3 0.3 1